[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M71FRX0TY5R9NMMRXJNN13V4":3,"skill-same:01M71FRX0TY5R9NMMRXJNN13V4":87},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":16,"sourceHtmlUrl":17,"categories":18,"keywords":24,"publishedAt":45,"linkOnly":46,"markdownContent":47,"markdownBody":48,"frontMatter":49,"refName":50,"commitSha":51,"skillDirectory":52,"forksCount":53,"repositoryHtmlUrl":54,"repositoryHomepageUrl":6,"ownerHtmlUrl":55,"ownerBlogUrl":56,"ownerDescription":57,"updatedAt":58,"install":59,"resources":72},"01M71FRX0TY5R9NMMRXJNN13V4","citation-check-skill",null,"Vision-enabled verification gate with web search. Use when users want to (1) verify slides\u002Freports\u002FPDFs\u002Fimages against authoritative online sources, (2) validate that citations actually exist and say what's claimed, (3) check charts\u002Fgraphs\u002Ftables for accuracy, (4) audit AI-generated content in doc-only mode (no external knowledge). Two modes - search mode validates against web, doc-only mode ensures everything traces to provided documents. Supports content in any language.","franklee16\u002Facademic-research-skills","franklee16","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F235809094?v=4","01M71ANES0C8SMYKF21NAE7YV0","Weikai Li","01M71ANESNNVKD3FDKBG01ZK5Y",234,"MIT","https:\u002F\u002Fgithub.com\u002Ffranklee16\u002Facademic-research-skills\u002Fblob\u002Fmaster\u002Freference-citation\u002Fcitation-check-skill-main\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Ffranklee16\u002Facademic-research-skills\u002Fblob\u002Fmaster\u002Freference-citation\u002Fcitation-check-skill-main\u002FSKILL.md",[19],{"categoryNo":20,"categoryName":21,"taxonomyType":22,"taxonomyTypeDesc":23},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44],"citation check skill","check","citation check","Citation","say","two","modes","accuracy","for","mode","tables","users","documents","use","verification gate","web search","slides","knowledge","Charts","search","2026-11-10 10:12:30",false,"---\nname: citation-check-skill\n\ndescription: Vision-enabled verification gate with web search. Use when users want to (1) verify slides\u002Freports\u002FPDFs\u002Fimages against authoritative online sources, (2) validate that citations actually exist and say what's claimed, (3) check charts\u002Fgraphs\u002Ftables for accuracy, (4) audit AI-generated content in doc-only mode (no external knowledge). Two modes - search mode validates against web, doc-only mode ensures everything traces to provided documents. Supports content in any language.\n\n---\n\n# Citation & Hallucination Checker v2\n\nVerification tool with vision + web search. Validates every claim against authoritative sources or provided documents. Works with content in any language.\n\n**Design principle:** Deterministic verification. Same input → Same output.\n\n---\n\n## Two Verification Modes\n\n### Mode 1: Search Verification (Default)\n\n- Searches web for authoritative sources\n- Validates citations actually exist\n- Checks if cited sources say what's claimed\n- Finds original data for statistics\n\n### Mode 2: Doc-Only Verification\n\n- User provides source document(s)\n- EVERYTHING must trace to those docs\n- Flags anything that appears to come from external knowledge\n- Trigger: \"only use this document\" \u002F \"verify against the PDF only\" \u002F \"don't search the web\"\n\n---\n\n## Two-Pass Architecture\n\n**Critical:** Always use two separate passes. Never interleave extraction and verification.\n\n### Pass 1: Extraction Only\n\n1. Read entire document\u002Fslides\u002Fimages\n2. Extract ALL claims using the Claim Extraction Rules below\n3. Output numbered list: `[claim_id] | [claim_text] | [claim_type] | [location]`\n4. **NO verification in this pass**\n5. Present extraction to user for confirmation before proceeding\n\n### Pass 2: Verification Only\n\n1. Take Pass 1 output as fixed input\n2. Verify each claim_id in sequential order\n3. **NO re-extraction allowed** — work only with Pass 1 claims\n4. Apply Status Decision Tree to each claim\n5. Generate final report\n\nThis prevents \"discovering new claims\" mid-verification and ensures consistency.\n\n---\n\n## Claim Extraction Rules (Exhaustive)\n\nExtract ONLY these claim types. Apply rules strictly — no judgment calls.\n\n### EXTRACT as claims:\n\n| Type            | Pattern                                                    | Example                                          |\n| --------------- | ---------------------------------------------------------- | ------------------------------------------------ |\n| **Statistic**   | Any number with unit\u002Fcontext (%, $, count, ratio, decimal) | \"92.3% accuracy\", \"$4.7B market\"                 |\n| **Comparative** | X is [comparative] than Y                                  | \"3x faster than baseline\"                        |\n| **Temporal**    | Time-bound assertion                                       | \"In 2024, adoption reached...\"                   |\n| **Attribution** | Claim tied to source                                       | \"According to WHO...\", \"Smith et al. found...\"   |\n| **Causal**      | X causes\u002Fleads to\u002Fresults in Y                             | \"This reduces latency by...\"                     |\n| **Existence**   | Asserts something exists\u002Fis true                           | \"There are 500M users\", \"The model supports...\"  |\n| **Ranking**     | Position claims                                            | \"largest\", \"first\", \"top 3\"                      |\n| **Quote**       | Direct quotation                                           | Any text in quotation marks attributed to source |\n\n### DO NOT extract as claims:\n\n| Type                              | Example                                            | Reason                          |\n| --------------------------------- | -------------------------------------------------- | ------------------------------- |\n| Definitions                       | \"Machine learning is a subset of AI\"               | Definitional, not factual claim |\n| Opinions marked as such           | \"We believe...\", \"In our view...\"                  | Explicitly subjective           |\n| Hypotheticals                     | \"If adoption continues...\", \"Could potentially...\" | Speculative                     |\n| Questions                         | \"What drives growth?\"                              | Not an assertion                |\n| Future predictions without source | \"Will reach $10B by 2030\"                          | Unless citing a forecast report |\n| Methodology descriptions          | \"We used PyTorch 2.0\"                              | Process, not factual claim      |\n| Acknowledgments                   | \"Thanks to our collaborators\"                      | Not verifiable                  |\n\n### Extraction Output Format\n\n```\n[C01] | \"Model achieves 96.555% accuracy on ImageNet\" | Statistic | Slide 3, bullet 2\n[C02] | \"Outperforms GPT-4 by 12% on reasoning tasks\" | Comparative | Slide 3, bullet 3\n[C03] | \"According to Chen et al. (2024), transformers scale linearly\" | Attribution | Slide 5, para 1\n[C04] | \"Market size reached $4.7B in 2024\" | Statistic + Temporal | Slide 7, chart title\n```\n\n---\n\n## Status Decision Tree\n\nApply this tree to EVERY claim. Follow exactly — no shortcuts.\n\n```\nSTART\n│\n├─ Is this a CITATION claim (references a paper\u002Freport\u002Fsource)?\n│   ├─ YES → Go to CITATION VALIDATION\n│   └─ NO → Go to STATISTIC\u002FFACT VALIDATION\n│\n│\nCITATION VALIDATION\n│\n├─ Step 1: Does the cited source exist?\n│   │   Run ALL mandatory search queries (see Search Templates)\n│   │\n│   ├─ NO → Status: \"Citation Not Found\"\n│   │        Issue: \"Cannot locate [citation] in any database\"\n│   │        STOP\n│   │\n│   └─ YES → Step 2: Does source contain the claimed topic?\n│             │\n│             ├─ NO → Status: \"Misquoted\"\n│             │        Issue: \"Source exists but does not discuss [topic]\"\n│             │        STOP\n│             │\n│             └─ YES → Step 3: Does source support the exact claim?\n│                       │\n│                       ├─ YES (exact match) → Status: \"Verified\"\n│                       │                       Confidence: \"exact\"\n│                       │\n│                       ├─ YES (paraphrase, same meaning) → Status: \"Verified\"\n│                       │                                   Confidence: \"paraphrase\"\n│                       │\n│                       ├─ PARTIALLY (missing context) → Status: \"Misleading\"\n│                       │   Issue: \"Claim omits critical context: [what's missing]\"\n│                       │\n│                       └─ NO (contradicts) → Status: \"Hallucination\"\n│                           Issue: \"Source says [X], claim says [Y]\"\n│\n│\nSTATISTIC\u002FFACT VALIDATION\n│\n├─ Step 1: Can you find an authoritative source?\n│   │   Run ALL mandatory search queries (see Search Templates)\n│   │\n│   ├─ NO (no source found) → Status: \"Unverified\"\n│   │                          Issue: \"No authoritative source found\"\n│   │                          STOP\n│   │\n│   └─ YES → Step 2: Do values match EXACTLY?\n│             │\n│             ├─ YES → Status: \"Verified\"\n│             │         Confidence: \"exact\"\n│             │         STOP\n│             │\n│             └─ NO → Status: \"Numerical Error\"\n│                      Go to NUMERICAL ERROR DETAILS\n│\n│\nNUMERICAL ERROR DETAILS (Academic Precision Mode)\n│\n├─ Record:\n│   • Source value: [exact number from source]\n│   • Claimed value: [number in document being checked]\n│   • Deviation: [calculate exact difference]\n│   • Source location: [page, table, section]\n│\n├─ Classification:\n│   • ANY rounding → Numerical Error\n│   • ANY truncation → Numerical Error  \n│   • Significant figures mismatch → Numerical Error\n│   • Unit mismatch → Numerical Error\n│   • Wrong direction (e.g., increase vs decrease) → Hallucination\n│\n└─ Exception: If source ITSELF provides rounded figure\n    • e.g., Source says \"96.555% (approximately 97%)\"\n    • Then claiming \"97%\" → Verified (cite the approximation)\n```\n\n---\n\n## Numerical Precision Rules (Academic Standard)\n\n**Default mode: Strict academic precision. Exact numbers only.**\n\n| Rule                 | Source      | Claim       | Status            |\n| -------------------- | ----------- | ----------- | ----------------- |\n| Exact match required | 96.555%     | 96.555%     | ✓ Verified        |\n| Any rounding = error | 96.555%     | 97%         | ✗ Numerical Error |\n| Any rounding = error | 96.555%     | 96.6%       | ✗ Numerical Error |\n| Truncation = error   | 96.555%     | 96.5%       | ✗ Numerical Error |\n| Sig figs must match  | 0.834       | 0.83        | ✗ Numerical Error |\n| Units must match     | 96.555%     | 0.96555     | ✗ Numerical Error |\n| Direction matters    | +12% growth | +15% growth | ✗ Hallucination   |\n| Order of magnitude   | $4.7B       | $47B        | ✗ Hallucination   |\n\n### Numerical Error Output Format\n\n```markdown\n### Numerical Error: [Claim ID]\n\n| Field | Value |\n|-------|-------|\n| Claim | \"Model achieves 97% accuracy\" |\n| Location | Slide 4, bullet 2 |\n| Source | Chen et al. (2024), Table 3, p.8 |\n| Source value | 96.555% |\n| Claimed value | 97% |\n| Deviation | +0.445% (rounded up) |\n| Status | Numerical Error |\n| Fix | Replace with: \"Model achieves 96.555% accuracy\" |\n```\n\n---\n\n## Confidence Classification\n\n| Level              | Criteria                                                   | Use when                            |\n| ------------------ | ---------------------------------------------------------- | ----------------------------------- |\n| **exact**          | ≥95% word overlap OR identical number with identical units | Direct quote, exact statistic       |\n| **paraphrase**     | Same fact, different words, no interpretation added        | Restated finding                    |\n| **interpretation** | Inference drawn from source data                           | Calculated from source, synthesized |\n\n**Rule:** When uncertain between levels, use the MORE CONSERVATIVE option and flag for review.\n\n---\n\n## Mandatory Search Templates\n\nRun ALL applicable templates. Do not stop after first result.\n\n### For Academic Citations\n\n```\nQuery 1: \"[first author last name] [year] [first 3 words of title]\"\nQuery 2: \"[full paper title]\" site:semanticscholar.org OR site:arxiv.org\nQuery 3: \"[first author] [year] [venue\u002Fjournal name]\"\nQuery 4: \"doi:[DOI]\" (if DOI provided)\nQuery 5: \"arxiv:[arxiv_id]\" (if arXiv ID provided)\n```\n\n### For Statistics (Market size, usage numbers, etc.)\n\n```\nQuery 1: \"[exact number with unit] [topic] [year]\"\nQuery 2: \"[topic] [year] statistics report site:statista.com\"\nQuery 3: \"[topic] [year] report site:mckinsey.com OR site:gartner.com\"\nQuery 4: \"[topic] market size [year] site:gov OR site:edu\"\nQuery 5: \"[topic] [number] original source\"\n```\n\n### For Company\u002FProduct Claims\n\n```\nQuery 1: \"[company name] [claim topic] press release [year]\"\nQuery 2: site:[company domain] [claim topic]\nQuery 3: \"[company name] [metric] official announcement\"\nQuery 4: \"[company name] [claim] SEC filing\" (for public companies)\n```\n\n### For Health\u002FMedical Claims\n\n```\nQuery 1: \"[claim topic] site:who.int OR site:cdc.gov OR site:nih.gov\"\nQuery 2: \"[claim] systematic review site:cochrane.org\"\nQuery 3: \"[claim] meta-analysis pubmed\"\n```\n\n### For Government\u002FPolicy Claims\n\n```\nQuery 1: \"[policy\u002Flaw name] site:gov\"\nQuery 2: \"[statistic] official statistics [country]\"\nQuery 3: \"[claim] [agency name] report\"\n```\n\n---\n\n## Source Authority Hierarchy\n\nWhen multiple sources found, prefer in this order:\n\n| Rank | Source Type                   | Examples                                          |\n| ---- | ----------------------------- | ------------------------------------------------- |\n| 1    | Primary source                | Original study, official report, raw data         |\n| 2    | Government\u002Finstitutional      | WHO, CDC, World Bank, national statistics offices |\n| 3    | Peer-reviewed publication     | Nature, Science, IEEE, ACM                        |\n| 4    | Industry reports (named)      | Gartner, McKinsey, Statista (with methodology)    |\n| 5    | Reputable news citing primary | NYT, Reuters citing original source               |\n| 6    | Secondary compilations        | Wikipedia (check their sources)                   |\n\n**Rule:** If only Rank 5-6 sources found, status = \"Unverified\" with note \"Only secondary sources found\"\n\n---\n\n## Multi-Source Verification (Search Mode)\n\nA claim achieves \"Verified\" status only if:\n\n| Condition              | Sources Required                                    |\n| ---------------------- | --------------------------------------------------- |\n| Primary source found   | 1 (if authoritative: .gov, peer-reviewed, official) |\n| Only secondary sources | ≥2 independent sources agreeing                     |\n| Sources conflict       | Status = \"Unverified\", note the conflict            |\n\n---\n\n## Tie-Breaker Rules\n\nWhen uncertain, apply these rules. No judgment calls.\n\n| Situation                                 | Rule                                                         |\n| ----------------------------------------- | ------------------------------------------------------------ |\n| Missing date on claim                     | Assume refers to most recent year available; flag \"needs date\" |\n| Conflicting sources                       | Use most recent authoritative source; cite both; note conflict |\n| Source not found after all queries        | Status = \"Unverified\" (NOT \"Hallucination\")                  |\n| Number differs due to currency conversion | Flag as \"Needs clarification: currency\u002Funits\"                |\n| Same org, multiple reports                | Use most recent; cite with date                              |\n| Claim uses \"approximately\" or \"about\"     | Still verify base number is in valid range (±10% of source)  |\n| Source is paywalled                       | Note \"Source behind paywall, unable to verify exact text\"    |\n| Source is in different language           | Translate and verify; note translation                       |\n\n---\n\n## Visual Data Verification\n\nFor every chart, graph, table, or diagram:\n\n### Step 1: Extract Data Points\n\n- Read ALL values from the visual\n- Record: axis labels, units, scale, legend\n- Note any visual distortions (truncated axes, 3D effects, etc.)\n\n### Step 2: Find Source\n\n- Search mode: Run search templates for the data\n- Doc-only mode: Locate in source document\n\n### Step 3: Compare Value-by-Value\n\n```markdown\n| Visual Element | Extracted Value | Source Value | Status |\n|----------------|-----------------|--------------|--------|\n| Bar 1 (2022) | 45% | 45.0% | ✓ Verified |\n| Bar 2 (2023) | 62% | 58.3% | ✗ Numerical Error |\n| Bar 3 (2024) | 78% | Not in source | ✗ Hallucination |\n```\n\n### Step 4: Check Visual Integrity\n\n| Check                                   | Issue Type                             |\n| --------------------------------------- | -------------------------------------- |\n| Y-axis starts at non-zero               | \"Visual Distortion: axis manipulation\" |\n| 3D effects distort proportions          | \"Visual Distortion: 3D exaggeration\"   |\n| Missing error bars when source has them | \"Misleading: uncertainty omitted\"      |\n| Different time ranges than source       | \"Misleading: cherry-picked timeframe\"  |\n\n---\n\n## Doc-Only Mode Workflow\n\n**Trigger phrases:**\n\n- \"only use this document\"\n- \"don't search the web\"\n- \"verify against the PDF only\"\n- \"everything should be from the source\"\n\n### Step 1: Index Source Document\n\nBuild complete index before any verification:\n\n```\nSOURCE INDEX\nDocument: [filename]\nPages: [count]\n\nPage 1:\n- Text: [summary of content]\n- Statistics: [list all numbers with context]\n- Tables: [Table 1: columns X, Y, Z]\n- Figures: [Figure 1: shows X]\n\nPage 2:\n...\n```\n\n### Step 2: Apply Two-Pass Architecture\n\nSame as search mode, but verification uses ONLY the source index.\n\n### Step 3: Trace Each Claim\n\n```\nClaim: [C01] \"Model achieves 92% accuracy\"\nSearch index for: \"92\", \"accuracy\", \"performance\"\n├─ Found: Section 4.2, p.8 — \"Our model achieves 92.1% accuracy\"\n│   └─ Status: Numerical Error (92% vs 92.1%)\n│\nOR\n│\n├─ Not found in index\n│   └─ Status: \"Not in Source\"\n│       Issue: \"This claim cannot be traced to the provided document\"\n│       Likely: External knowledge \u002F hallucination\n```\n\n### Step 4: Flag ALL External Knowledge\n\nIn doc-only mode, ANY claim not traceable to source = problem\n\n```markdown\n### External Knowledge Detected\n\nThese claims are NOT in the provided document:\n\n| Claim ID | Claim | Status | Issue |\n|----------|-------|--------|-------|\n| C07 | \"This method is widely adopted in industry\" | Not in Source | Appears to be from model training data |\n| C12 | \"Published in Nature 2024\" | Not in Source | Publication venue not mentioned in source |\n```\n\n---\n\n## Output Format\n\n### Summary Block (Always First)\n\n```markdown\n## Verification Report\n\n**Mode:** [Search \u002F Doc-Only]\n**Document:** [filename or description]\n**Generated:** [timestamp]\n\n### Summary\n| Metric | Count |\n|--------|-------|\n| Total claims extracted | X |\n| Verified | Y |\n| Numerical Error | Z |\n| Unverified | A |\n| Hallucination | B |\n| Misleading | C |\n| Not in Source (doc-only) | D |\n\n**Overall Status:** [PASS: All verified \u002F FAIL: Issues found]\n```\n\n### Detailed Findings (Grouped by Status)\n\n```markdown\n### ✓ Verified Claims (N)\n\n| ID | Claim | Source | Location | Confidence |\n|----|-------|--------|----------|------------|\n| C01 | \"92.1% accuracy\" | Chen et al. 2024 | Table 3, p.8 | exact |\n\n### ✗ Numerical Errors (N)\n\n| ID | Claim | Source Value | Claimed Value | Deviation | Fix |\n|----|-------|--------------|---------------|-----------|-----|\n| C03 | \"97% accuracy\" | 96.555% | 97% | +0.445% | Use 96.555% |\n\n### ✗ Hallucinations (N)\n\n| ID | Claim | Issue | Source Says |\n|----|-------|-------|-------------|\n| C05 | \"3x faster\" | Contradicts source | Source: 2.1x faster |\n\n### ⚠ Unverified (N)\n\n| ID | Claim | Issue |\n|----|-------|-------|\n| C08 | \"$50B market\" | No authoritative source found |\n\n### ⚠ Misleading (N)\n\n| ID | Claim | Issue | Missing Context |\n|----|-------|-------|-----------------|\n| C10 | \"Best performance\" | Cherry-picked metric | Only on subset; overall performance lower |\n```\n\n### Sources Consulted\n\n```markdown\n### Sources\n\n| ID | Citation | Type | URL | Used For |\n|----|----------|------|-----|----------|\n| S1 | Chen et al. (2024) | arxiv | https:\u002F\u002Farxiv.org\u002F... | C01, C02, C03 |\n| S2 | Statista Market Report | report | https:\u002F\u002Fstatista.com\u002F... | C08 |\n```\n\n---\n\n## Critical Rules\n\n1. **Two-pass always** — Extract first, verify second. Never interleave.\n2. **Every claim gets checked** — No exceptions, no skipping \"obvious\" ones.\n3. **Exact numbers only** — 96.555% ≠ 97% in academic mode.\n4. **Find the origin** — Don't accept secondary sources citing unknown primaries.\n5. **Run ALL search templates** — Don't stop after first result.\n6. **Citations must be real** — Search to confirm papers\u002Freports exist.\n7. **Check what sources actually say** — A real paper can still be misquoted.\n8. **In doc-only mode, flag ALL external knowledge** — Even if it's true.\n9. **When uncertain, be conservative** — \"Unverified\" is safer than false \"Verified\".\n10. **Follow tie-breaker rules** — No ad-hoc judgment calls.\n\n---\n\n## Language Support\n\n- Accepts content in any language\n- Searches in the appropriate language for sources\n- Reports in the same language as user's request\n- Cross-language verification supported (e.g., Chinese slides citing English papers)\n- When translating for verification, note: \"Translated from [language]\"\n\n---\n\n## Output Format Options\n\n**Quick:** Summary + critical issues only (Numerical Errors, Hallucinations, Unverified)\n**Full:** Complete traceability report with all claims\n**JSON:** Machine-readable audit (see references\u002Fcitation_schema.json)\n\n---\n\n## Changelog\n\n**v2.0** — Consistency update\n\n- Added Two-Pass Architecture (extract → verify separation)\n- Added exhaustive Claim Extraction Rules\n- Added Status Decision Tree (deterministic classification)\n- Added strict Numerical Precision Rules (academic mode)\n- Added Mandatory Search Templates\n- Added Multi-Source Verification requirements\n- Added Tie-Breaker Rules for edge cases\n- Added Confidence Classification thresholds\n","# Citation & Hallucination Checker v2\n\nVerification tool with vision + web search. Validates every claim against authoritative sources or provided documents. Works with content in any language.\n\n**Design principle:** Deterministic verification. Same input → Same output.\n\n---\n\n## Two Verification Modes\n\n### Mode 1: Search Verification (Default)\n\n- Searches web for authoritative sources\n- Validates citations actually exist\n- Checks if cited sources say what's claimed\n- Finds original data for statistics\n\n### Mode 2: Doc-Only Verification\n\n- User provides source document(s)\n- EVERYTHING must trace to those docs\n- Flags anything that appears to come from external knowledge\n- Trigger: \"only use this document\" \u002F \"verify against the PDF only\" \u002F \"don't search the web\"\n\n---\n\n## Two-Pass Architecture\n\n**Critical:** Always use two separate passes. Never interleave extraction and verification.\n\n### Pass 1: Extraction Only\n\n1. Read entire document\u002Fslides\u002Fimages\n2. Extract ALL claims using the Claim Extraction Rules below\n3. Output numbered list: `[claim_id] | [claim_text] | [claim_type] | [location]`\n4. **NO verification in this pass**\n5. Present extraction to user for confirmation before proceeding\n\n### Pass 2: Verification Only\n\n1. Take Pass 1 output as fixed input\n2. Verify each claim_id in sequential order\n3. **NO re-extraction allowed** — work only with Pass 1 claims\n4. Apply Status Decision Tree to each claim\n5. Generate final report\n\nThis prevents \"discovering new claims\" mid-verification and ensures consistency.\n\n---\n\n## Claim Extraction Rules (Exhaustive)\n\nExtract ONLY these claim types. Apply rules strictly — no judgment calls.\n\n### EXTRACT as claims:\n\n| Type            | Pattern                                                    | Example                                          |\n| --------------- | ---------------------------------------------------------- | ------------------------------------------------ |\n| **Statistic**   | Any number with unit\u002Fcontext (%, $, count, ratio, decimal) | \"92.3% accuracy\", \"$4.7B market\"                 |\n| **Comparative** | X is [comparative] than Y                                  | \"3x faster than baseline\"                        |\n| **Temporal**    | Time-bound assertion                                       | \"In 2024, adoption reached...\"                   |\n| **Attribution** | Claim tied to source                                       | \"According to WHO...\", \"Smith et al. found...\"   |\n| **Causal**      | X causes\u002Fleads to\u002Fresults in Y                             | \"This reduces latency by...\"                     |\n| **Existence**   | Asserts something exists\u002Fis true                           | \"There are 500M users\", \"The model supports...\"  |\n| **Ranking**     | Position claims                                            | \"largest\", \"first\", \"top 3\"                      |\n| **Quote**       | Direct quotation                                           | Any text in quotation marks attributed to source |\n\n### DO NOT extract as claims:\n\n| Type                              | Example                                            | Reason                          |\n| --------------------------------- | -------------------------------------------------- | ------------------------------- |\n| Definitions                       | \"Machine learning is a subset of AI\"               | Definitional, not factual claim |\n| Opinions marked as such           | \"We believe...\", \"In our view...\"                  | Explicitly subjective           |\n| Hypotheticals                     | \"If adoption continues...\", \"Could potentially...\" | Speculative                     |\n| Questions                         | \"What drives growth?\"                              | Not an assertion                |\n| Future predictions without source | \"Will reach $10B by 2030\"                          | Unless citing a forecast report |\n| Methodology descriptions          | \"We used PyTorch 2.0\"                              | Process, not factual claim      |\n| Acknowledgments                   | \"Thanks to our collaborators\"                      | Not verifiable                  |\n\n### Extraction Output Format\n\n```\n[C01] | \"Model achieves 96.555% accuracy on ImageNet\" | Statistic | Slide 3, bullet 2\n[C02] | \"Outperforms GPT-4 by 12% on reasoning tasks\" | Comparative | Slide 3, bullet 3\n[C03] | \"According to Chen et al. (2024), transformers scale linearly\" | Attribution | Slide 5, para 1\n[C04] | \"Market size reached $4.7B in 2024\" | Statistic + Temporal | Slide 7, chart title\n```\n\n---\n\n## Status Decision Tree\n\nApply this tree to EVERY claim. Follow exactly — no shortcuts.\n\n```\nSTART\n│\n├─ Is this a CITATION claim (references a paper\u002Freport\u002Fsource)?\n│   ├─ YES → Go to CITATION VALIDATION\n│   └─ NO → Go to STATISTIC\u002FFACT VALIDATION\n│\n│\nCITATION VALIDATION\n│\n├─ Step 1: Does the cited source exist?\n│   │   Run ALL mandatory search queries (see Search Templates)\n│   │\n│   ├─ NO → Status: \"Citation Not Found\"\n│   │        Issue: \"Cannot locate [citation] in any database\"\n│   │        STOP\n│   │\n│   └─ YES → Step 2: Does source contain the claimed topic?\n│             │\n│             ├─ NO → Status: \"Misquoted\"\n│             │        Issue: \"Source exists but does not discuss [topic]\"\n│             │        STOP\n│             │\n│             └─ YES → Step 3: Does source support the exact claim?\n│                       │\n│                       ├─ YES (exact match) → Status: \"Verified\"\n│                       │                       Confidence: \"exact\"\n│                       │\n│                       ├─ YES (paraphrase, same meaning) → Status: \"Verified\"\n│                       │                                   Confidence: \"paraphrase\"\n│                       │\n│                       ├─ PARTIALLY (missing context) → Status: \"Misleading\"\n│                       │   Issue: \"Claim omits critical context: [what's missing]\"\n│                       │\n│                       └─ NO (contradicts) → Status: \"Hallucination\"\n│                           Issue: \"Source says [X], claim says [Y]\"\n│\n│\nSTATISTIC\u002FFACT VALIDATION\n│\n├─ Step 1: Can you find an authoritative source?\n│   │   Run ALL mandatory search queries (see Search Templates)\n│   │\n│   ├─ NO (no source found) → Status: \"Unverified\"\n│   │                          Issue: \"No authoritative source found\"\n│   │                          STOP\n│   │\n│   └─ YES → Step 2: Do values match EXACTLY?\n│             │\n│             ├─ YES → Status: \"Verified\"\n│             │         Confidence: \"exact\"\n│             │         STOP\n│             │\n│             └─ NO → Status: \"Numerical Error\"\n│                      Go to NUMERICAL ERROR DETAILS\n│\n│\nNUMERICAL ERROR DETAILS (Academic Precision Mode)\n│\n├─ Record:\n│   • Source value: [exact number from source]\n│   • Claimed value: [number in document being checked]\n│   • Deviation: [calculate exact difference]\n│   • Source location: [page, table, section]\n│\n├─ Classification:\n│   • ANY rounding → Numerical Error\n│   • ANY truncation → Numerical Error  \n│   • Significant figures mismatch → Numerical Error\n│   • Unit mismatch → Numerical Error\n│   • Wrong direction (e.g., increase vs decrease) → Hallucination\n│\n└─ Exception: If source ITSELF provides rounded figure\n    • e.g., Source says \"96.555% (approximately 97%)\"\n    • Then claiming \"97%\" → Verified (cite the approximation)\n```\n\n---\n\n## Numerical Precision Rules (Academic Standard)\n\n**Default mode: Strict academic precision. Exact numbers only.**\n\n| Rule                 | Source      | Claim       | Status            |\n| -------------------- | ----------- | ----------- | ----------------- |\n| Exact match required | 96.555%     | 96.555%     | ✓ Verified        |\n| Any rounding = error | 96.555%     | 97%         | ✗ Numerical Error |\n| Any rounding = error | 96.555%     | 96.6%       | ✗ Numerical Error |\n| Truncation = error   | 96.555%     | 96.5%       | ✗ Numerical Error |\n| Sig figs must match  | 0.834       | 0.83        | ✗ Numerical Error |\n| Units must match     | 96.555%     | 0.96555     | ✗ Numerical Error |\n| Direction matters    | +12% growth | +15% growth | ✗ Hallucination   |\n| Order of magnitude   | $4.7B       | $47B        | ✗ Hallucination   |\n\n### Numerical Error Output Format\n\n```markdown\n### Numerical Error: [Claim ID]\n\n| Field | Value |\n|-------|-------|\n| Claim | \"Model achieves 97% accuracy\" |\n| Location | Slide 4, bullet 2 |\n| Source | Chen et al. (2024), Table 3, p.8 |\n| Source value | 96.555% |\n| Claimed value | 97% |\n| Deviation | +0.445% (rounded up) |\n| Status | Numerical Error |\n| Fix | Replace with: \"Model achieves 96.555% accuracy\" |\n```\n\n---\n\n## Confidence Classification\n\n| Level              | Criteria                                                   | Use when                            |\n| ------------------ | ---------------------------------------------------------- | ----------------------------------- |\n| **exact**          | ≥95% word overlap OR identical number with identical units | Direct quote, exact statistic       |\n| **paraphrase**     | Same fact, different words, no interpretation added        | Restated finding                    |\n| **interpretation** | Inference drawn from source data                           | Calculated from source, synthesized |\n\n**Rule:** When uncertain between levels, use the MORE CONSERVATIVE option and flag for review.\n\n---\n\n## Mandatory Search Templates\n\nRun ALL applicable templates. Do not stop after first result.\n\n### For Academic Citations\n\n```\nQuery 1: \"[first author last name] [year] [first 3 words of title]\"\nQuery 2: \"[full paper title]\" site:semanticscholar.org OR site:arxiv.org\nQuery 3: \"[first author] [year] [venue\u002Fjournal name]\"\nQuery 4: \"doi:[DOI]\" (if DOI provided)\nQuery 5: \"arxiv:[arxiv_id]\" (if arXiv ID provided)\n```\n\n### For Statistics (Market size, usage numbers, etc.)\n\n```\nQuery 1: \"[exact number with unit] [topic] [year]\"\nQuery 2: \"[topic] [year] statistics report site:statista.com\"\nQuery 3: \"[topic] [year] report site:mckinsey.com OR site:gartner.com\"\nQuery 4: \"[topic] market size [year] site:gov OR site:edu\"\nQuery 5: \"[topic] [number] original source\"\n```\n\n### For Company\u002FProduct Claims\n\n```\nQuery 1: \"[company name] [claim topic] press release [year]\"\nQuery 2: site:[company domain] [claim topic]\nQuery 3: \"[company name] [metric] official announcement\"\nQuery 4: \"[company name] [claim] SEC filing\" (for public companies)\n```\n\n### For Health\u002FMedical Claims\n\n```\nQuery 1: \"[claim topic] site:who.int OR site:cdc.gov OR site:nih.gov\"\nQuery 2: \"[claim] systematic review site:cochrane.org\"\nQuery 3: \"[claim] meta-analysis pubmed\"\n```\n\n### For Government\u002FPolicy Claims\n\n```\nQuery 1: \"[policy\u002Flaw name] site:gov\"\nQuery 2: \"[statistic] official statistics [country]\"\nQuery 3: \"[claim] [agency name] report\"\n```\n\n---\n\n## Source Authority Hierarchy\n\nWhen multiple sources found, prefer in this order:\n\n| Rank | Source Type                   | Examples                                          |\n| ---- | ----------------------------- | ------------------------------------------------- |\n| 1    | Primary source                | Original study, official report, raw data         |\n| 2    | Government\u002Finstitutional      | WHO, CDC, World Bank, national statistics offices |\n| 3    | Peer-reviewed publication     | Nature, Science, IEEE, ACM                        |\n| 4    | Industry reports (named)      | Gartner, McKinsey, Statista (with methodology)    |\n| 5    | Reputable news citing primary | NYT, Reuters citing original source               |\n| 6    | Secondary compilations        | Wikipedia (check their sources)                   |\n\n**Rule:** If only Rank 5-6 sources found, status = \"Unverified\" with note \"Only secondary sources found\"\n\n---\n\n## Multi-Source Verification (Search Mode)\n\nA claim achieves \"Verified\" status only if:\n\n| Condition              | Sources Required                                    |\n| ---------------------- | --------------------------------------------------- |\n| Primary source found   | 1 (if authoritative: .gov, peer-reviewed, official) |\n| Only secondary sources | ≥2 independent sources agreeing                     |\n| Sources conflict       | Status = \"Unverified\", note the conflict            |\n\n---\n\n## Tie-Breaker Rules\n\nWhen uncertain, apply these rules. No judgment calls.\n\n| Situation                                 | Rule                                                         |\n| ----------------------------------------- | ------------------------------------------------------------ |\n| Missing date on claim                     | Assume refers to most recent year available; flag \"needs date\" |\n| Conflicting sources                       | Use most recent authoritative source; cite both; note conflict |\n| Source not found after all queries        | Status = \"Unverified\" (NOT \"Hallucination\")                  |\n| Number differs due to currency conversion | Flag as \"Needs clarification: currency\u002Funits\"                |\n| Same org, multiple reports                | Use most recent; cite with date                              |\n| Claim uses \"approximately\" or \"about\"     | Still verify base number is in valid range (±10% of source)  |\n| Source is paywalled                       | Note \"Source behind paywall, unable to verify exact text\"    |\n| Source is in different language           | Translate and verify; note translation                       |\n\n---\n\n## Visual Data Verification\n\nFor every chart, graph, table, or diagram:\n\n### Step 1: Extract Data Points\n\n- Read ALL values from the visual\n- Record: axis labels, units, scale, legend\n- Note any visual distortions (truncated axes, 3D effects, etc.)\n\n### Step 2: Find Source\n\n- Search mode: Run search templates for the data\n- Doc-only mode: Locate in source document\n\n### Step 3: Compare Value-by-Value\n\n```markdown\n| Visual Element | Extracted Value | Source Value | Status |\n|----------------|-----------------|--------------|--------|\n| Bar 1 (2022) | 45% | 45.0% | ✓ Verified |\n| Bar 2 (2023) | 62% | 58.3% | ✗ Numerical Error |\n| Bar 3 (2024) | 78% | Not in source | ✗ Hallucination |\n```\n\n### Step 4: Check Visual Integrity\n\n| Check                                   | Issue Type                             |\n| --------------------------------------- | -------------------------------------- |\n| Y-axis starts at non-zero               | \"Visual Distortion: axis manipulation\" |\n| 3D effects distort proportions          | \"Visual Distortion: 3D exaggeration\"   |\n| Missing error bars when source has them | \"Misleading: uncertainty omitted\"      |\n| Different time ranges than source       | \"Misleading: cherry-picked timeframe\"  |\n\n---\n\n## Doc-Only Mode Workflow\n\n**Trigger phrases:**\n\n- \"only use this document\"\n- \"don't search the web\"\n- \"verify against the PDF only\"\n- \"everything should be from the source\"\n\n### Step 1: Index Source Document\n\nBuild complete index before any verification:\n\n```\nSOURCE INDEX\nDocument: [filename]\nPages: [count]\n\nPage 1:\n- Text: [summary of content]\n- Statistics: [list all numbers with context]\n- Tables: [Table 1: columns X, Y, Z]\n- Figures: [Figure 1: shows X]\n\nPage 2:\n...\n```\n\n### Step 2: Apply Two-Pass Architecture\n\nSame as search mode, but verification uses ONLY the source index.\n\n### Step 3: Trace Each Claim\n\n```\nClaim: [C01] \"Model achieves 92% accuracy\"\nSearch index for: \"92\", \"accuracy\", \"performance\"\n├─ Found: Section 4.2, p.8 — \"Our model achieves 92.1% accuracy\"\n│   └─ Status: Numerical Error (92% vs 92.1%)\n│\nOR\n│\n├─ Not found in index\n│   └─ Status: \"Not in Source\"\n│       Issue: \"This claim cannot be traced to the provided document\"\n│       Likely: External knowledge \u002F hallucination\n```\n\n### Step 4: Flag ALL External Knowledge\n\nIn doc-only mode, ANY claim not traceable to source = problem\n\n```markdown\n### External Knowledge Detected\n\nThese claims are NOT in the provided document:\n\n| Claim ID | Claim | Status | Issue |\n|----------|-------|--------|-------|\n| C07 | \"This method is widely adopted in industry\" | Not in Source | Appears to be from model training data |\n| C12 | \"Published in Nature 2024\" | Not in Source | Publication venue not mentioned in source |\n```\n\n---\n\n## Output Format\n\n### Summary Block (Always First)\n\n```markdown\n## Verification Report\n\n**Mode:** [Search \u002F Doc-Only]\n**Document:** [filename or description]\n**Generated:** [timestamp]\n\n### Summary\n| Metric | Count |\n|--------|-------|\n| Total claims extracted | X |\n| Verified | Y |\n| Numerical Error | Z |\n| Unverified | A |\n| Hallucination | B |\n| Misleading | C |\n| Not in Source (doc-only) | D |\n\n**Overall Status:** [PASS: All verified \u002F FAIL: Issues found]\n```\n\n### Detailed Findings (Grouped by Status)\n\n```markdown\n### ✓ Verified Claims (N)\n\n| ID | Claim | Source | Location | Confidence |\n|----|-------|--------|----------|------------|\n| C01 | \"92.1% accuracy\" | Chen et al. 2024 | Table 3, p.8 | exact |\n\n### ✗ Numerical Errors (N)\n\n| ID | Claim | Source Value | Claimed Value | Deviation | Fix |\n|----|-------|--------------|---------------|-----------|-----|\n| C03 | \"97% accuracy\" | 96.555% | 97% | +0.445% | Use 96.555% |\n\n### ✗ Hallucinations (N)\n\n| ID | Claim | Issue | Source Says |\n|----|-------|-------|-------------|\n| C05 | \"3x faster\" | Contradicts source | Source: 2.1x faster |\n\n### ⚠ Unverified (N)\n\n| ID | Claim | Issue |\n|----|-------|-------|\n| C08 | \"$50B market\" | No authoritative source found |\n\n### ⚠ Misleading (N)\n\n| ID | Claim | Issue | Missing Context |\n|----|-------|-------|-----------------|\n| C10 | \"Best performance\" | Cherry-picked metric | Only on subset; overall performance lower |\n```\n\n### Sources Consulted\n\n```markdown\n### Sources\n\n| ID | Citation | Type | URL | Used For |\n|----|----------|------|-----|----------|\n| S1 | Chen et al. (2024) | arxiv | https:\u002F\u002Farxiv.org\u002F... | C01, C02, C03 |\n| S2 | Statista Market Report | report | https:\u002F\u002Fstatista.com\u002F... | C08 |\n```\n\n---\n\n## Critical Rules\n\n1. **Two-pass always** — Extract first, verify second. Never interleave.\n2. **Every claim gets checked** — No exceptions, no skipping \"obvious\" ones.\n3. **Exact numbers only** — 96.555% ≠ 97% in academic mode.\n4. **Find the origin** — Don't accept secondary sources citing unknown primaries.\n5. **Run ALL search templates** — Don't stop after first result.\n6. **Citations must be real** — Search to confirm papers\u002Freports exist.\n7. **Check what sources actually say** — A real paper can still be misquoted.\n8. **In doc-only mode, flag ALL external knowledge** — Even if it's true.\n9. **When uncertain, be conservative** — \"Unverified\" is safer than false \"Verified\".\n10. **Follow tie-breaker rules** — No ad-hoc judgment calls.\n\n---\n\n## Language Support\n\n- Accepts content in any language\n- Searches in the appropriate language for sources\n- Reports in the same language as user's request\n- Cross-language verification supported (e.g., Chinese slides citing English papers)\n- When translating for verification, note: \"Translated from [language]\"\n\n---\n\n## Output Format Options\n\n**Quick:** Summary + critical issues only (Numerical Errors, Hallucinations, Unverified)\n**Full:** Complete traceability report with all claims\n**JSON:** Machine-readable audit (see references\u002Fcitation_schema.json)\n\n---\n\n## Changelog\n\n**v2.0** — Consistency update\n\n- Added Two-Pass Architecture (extract → verify separation)\n- Added exhaustive Claim Extraction Rules\n- Added Status Decision Tree (deterministic classification)\n- Added strict Numerical Precision Rules (academic mode)\n- Added Mandatory Search Templates\n- Added Multi-Source Verification requirements\n- Added Tie-Breaker Rules for edge cases\n- Added Confidence Classification thresholds\n",{"name":5,"description":7},"master","9a4b2db21fc4b217136ae4a0e6faad42d179ca03","reference-citation\u002Fcitation-check-skill-main",36,"https:\u002F\u002Fgithub.com\u002Ffranklee16\u002Facademic-research-skills","https:\u002F\u002Fgithub.com\u002Ffranklee16","https:\u002F\u002Fsites.google.com\u002Fview\u002Ffrankweikaili\u002Fhome","Associate Professor of Finance at City University of Hong Kong - empirical asset pricing, behavioral finance, ESG. I build AI tools for research and teaching","2026-09-18 20:12:00",{"source":8,"skillName":5,"commands":60,"prompt":70,"zipAvailable":71},[61,64,67],{"tool":62,"command":63},"npx","npx skills add franklee16\u002Facademic-research-skills --skill citation-check-skill",{"tool":65,"command":66},"bunx","bunx skills add franklee16\u002Facademic-research-skills --skill citation-check-skill",{"tool":68,"command":69},"pnpm","pnpm dlx skills add franklee16\u002Facademic-research-skills --skill citation-check-skill","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Ffranklee16\u002Facademic-research-skills\u002Fblob\u002Fmaster\u002Freference-citation\u002Fcitation-check-skill-main\u002FSKILL.md",true,[73,77,83],{"path":74,"resourceType":74,"resourceTypeDesc":75,"sizeBytes":76,"sourceHtmlUrl":16},"LICENSE","许可证",1063,{"path":78,"resourceType":79,"resourceTypeDesc":80,"sizeBytes":81,"sourceHtmlUrl":82},"README.md","REFERENCE","参考文档",4360,"https:\u002F\u002Fgithub.com\u002Ffranklee16\u002Facademic-research-skills\u002Fblob\u002Fmaster\u002Freference-citation\u002Fcitation-check-skill-main\u002FREADME.md",{"path":84,"resourceType":79,"resourceTypeDesc":80,"sizeBytes":85,"sourceHtmlUrl":86},"references\u002Fcitation_schema.json",4821,"https:\u002F\u002Fgithub.com\u002Ffranklee16\u002Facademic-research-skills\u002Fblob\u002Fmaster\u002Freference-citation\u002Fcitation-check-skill-main\u002Freferences\u002Fcitation_schema.json",[88,124,153,182,207,228],{"skillNo":89,"name":90,"nameZh":6,"description":91,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":92,"sourceHtmlUrl":93,"categories":94,"keywords":103,"publishedAt":45},"01M71G52NMZZKJ7BRCJDNVXECV","thu-digitizer","Extract and validate numeric data from chart images, PDF figures, and official source-data files locally and reproducibly, with calibrated axes, compact-scatter recovery plus residual audits, grouped\u002Fstacked bars, visible-label pie\u002Fdonut validation, original-pixel aligned lattice composites such as UpSet plots, PDF vector inspection, line tracing, histograms, boxplots, error bars, CSV\u002FJSON evidence, source-data cross-validation, and regression evaluation. Use when a user asks to recover or recreate chart data, extract scatter points, bars, printed pie\u002Fdonut labels, or aligned memberships, check a PDF chart, assess digitization accuracy, validate against source data, or improve a chart-digitization workflow from corrected examples.","https:\u002F\u002Fgithub.com\u002Ffranklee16\u002Facademic-research-skills\u002Fblob\u002Fmaster\u002Fvisualization\u002Fthu-digitizer-main\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Ffranklee16\u002Facademic-research-skills\u002Fblob\u002Fmaster\u002Fvisualization\u002Fthu-digitizer-main\u002FSKILL.md",[95,98],{"categoryNo":96,"categoryName":97,"taxonomyType":22,"taxonomyTypeDesc":23},"01JBSEEDCAT000000000000002","文档处理",{"categoryNo":99,"categoryName":100,"taxonomyType":101,"taxonomyTypeDesc":102},"01JBSEEDCAT000000000000033","数据分析师","PROFESSION","适用职业",[104,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119,120,121,122,123],"thu digitizer","cross-validation","line","Evaluation","workflow","data","inspection","regression","official","labels","pixel","improve","files","PDF","error","images","vector","chart","compact","when",{"skillNo":125,"name":126,"nameZh":6,"description":127,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":128,"categories":129,"keywords":134,"publishedAt":45},"01M71G4XJWMJST9K8X244GXH71","scientific-visualization","Meta-skill for publication-ready figures. Use when creating journal submission figures requiring multi-panel layouts, significance annotations, error bars, colorblind-safe palettes, and specific journal formatting (Nature, Science, Cell). Orchestrates matplotlib\u002Fseaborn\u002Fplotly with publication styles. For quick exploration use seaborn or plotly directly.","https:\u002F\u002Fgithub.com\u002Ffranklee16\u002Facademic-research-skills\u002Fblob\u002Fmaster\u002Fvisualization\u002Fscientific-visualization\u002FSKILL.md",[130,133],{"categoryNo":131,"categoryName":132,"taxonomyType":22,"taxonomyTypeDesc":23},"01JBSEEDCAT000000000000013","分析与可视化",{"categoryNo":20,"categoryName":21,"taxonomyType":22,"taxonomyTypeDesc":23},[135,136,137,138,139,140,141,123,142,143,144,145,146,147,118,148,149,150,151,152],"scientific visualization","scientific","visualization","Meta","plotly","safe","journal submission","journal","meta-skill","exploration","quick","nature","ready","panel","figures","Publication","formatting","annotations",{"skillNo":154,"name":155,"nameZh":6,"description":156,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":157,"sourceHtmlUrl":158,"categories":159,"keywords":164,"publishedAt":45},"01M71G3P8C430F3J0XX152ZJ4F","theoretical-economics-claude-skill","Human-in-the-loop theoretical economics research pipeline: from economic intuition to manuscript skeleton in 10 structured stages with quality gates.","https:\u002F\u002Fgithub.com\u002Ffranklee16\u002Facademic-research-skills\u002Fblob\u002Fmaster\u002Ftheory-skills\u002FpAI-Econ-claude-main\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Ffranklee16\u002Facademic-research-skills\u002Fblob\u002Fmaster\u002Ftheory-skills\u002FpAI-Econ-claude-main\u002FSKILL.md",[160,163],{"categoryNo":161,"categoryName":162,"taxonomyType":22,"taxonomyTypeDesc":23},"01JBSEEDCAT000000000000021","资料检索与研究",{"categoryNo":20,"categoryName":21,"taxonomyType":22,"taxonomyTypeDesc":23},[165,166,167,168,169,170,171,172,173,174,175,176,177,178,179,180,181],"theoretical economics claude skill","economics","claude skill","human","quality","research","quality gates","with","gates","true","human-in-the-loop","manuscript","research pipeline","structured","loop","skeleton","pipeline",{"skillNo":183,"name":184,"nameZh":6,"description":185,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":186,"sourceHtmlUrl":187,"categories":188,"keywords":199,"publishedAt":45},"01M71FZF5ZB0384VC60EKQ885N","interview-guide-design","把一个研究题目、研究思路或研究问题，落成一份能用于学术研究、案例研究、学位论文、课题调研的质性访谈提纲时，使用此 skill。触发信号很宽：用户说\"做\u002F设计访谈提纲\"\"访谈大纲怎么设计才规范\"\"结构性访谈提纲\"\"把研究问题\u002F题目变成访谈问题\"\"帮我弄份访谈提纲去问问他们\"\"怎么访谈才能挖到深层的东西\"\"interview guide \u002F interview protocol \u002F key informant \u002F elite interview\"，无论措辞多口语、有没有提方法论，都应触发。典型情形：要去访谈领导 \u002F 干部 \u002F 管理者 \u002F 专家 \u002F 科室主任 \u002F 一线员工或老师 \u002F 关键知情人，了解某个现象、做法或机制；半结构化 \u002F 结构化 \u002F 关键知情人 \u002F 精英访谈皆可；中英双语；社科 \u002F 管理 \u002F 教育 \u002F 公共政策 \u002F 医疗等领域；也常与案例研究、或与一份配套问卷的混合方法调研一起出现。它把研究问题按 Castillo-Montoya 的 IPR 框架与 Yin\u002FStake 的案例研究规范，拆成\"主问题—追问—探测\"，保障与研究问题对齐、可信度（trustworthiness）与三角验证就绪，标注方法论依据与权威出处，产出研究者版 + 现场详版\u002F简版。不要用于：量化问卷 \u002F 量表设计（走 questionnaire-design）、求职面试、以传播或叙事为目的的人物访谈（媒体\u002F播客\u002F记者深度报道\u002F纪录片\u002F口述史，非以回答研究问题为目的）、访谈录音转写、焦点小组\u002F座谈会提纲（群体互动动态与主持伦理不同，本 skill 不覆盖）、以及访谈完成后的编码与主题分析。","https:\u002F\u002Fgithub.com\u002Ffranklee16\u002Facademic-research-skills\u002Fblob\u002Fmaster\u002Frunning-surveys-experiments\u002Fmixed-methods-instrument-design-main\u002Freferences\u002Finterview-guide-design\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Ffranklee16\u002Facademic-research-skills\u002Fblob\u002Fmaster\u002Frunning-surveys-experiments\u002Fmixed-methods-instrument-design-main\u002Freferences\u002Finterview-guide-design\u002FSKILL.md",[189,192,193,196],{"categoryNo":190,"categoryName":191,"taxonomyType":22,"taxonomyTypeDesc":23},"01JBSEEDCAT000000000000015","视觉设计",{"categoryNo":161,"categoryName":162,"taxonomyType":22,"taxonomyTypeDesc":23},{"categoryNo":194,"categoryName":195,"taxonomyType":101,"taxonomyTypeDesc":102},"01JBSEEDCAT000000000000032","设计师",{"categoryNo":197,"categoryName":198,"taxonomyType":101,"taxonomyTypeDesc":102},"01JBSEEDCAT000000000000038","人力资源",[200,201,202,203,204,205,206],"interview guide design","interview","interview guide","design","guide","questionnaire","protocol",{"skillNo":208,"name":209,"nameZh":6,"description":210,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":211,"sourceHtmlUrl":212,"categories":213,"keywords":218,"publishedAt":45},"01M71FZBQ29K2JPYZD1HWWTZC9","mixed-methods-instrument-design","基于同一项研究的研究目的、研究问题、理论框架与人群设计，协同产出彼此对齐的量化问卷\u002F调查表和质性访谈提纲时使用；也可在用户明确要求时把经审核的问卷创建为问卷星未发布草稿并继续管理。触发信号包括“问卷和访谈一起设计”“根据研究思路同步出调查问卷与访谈提纲”“混合方法研究工具”“不同人群分别做问卷和访谈”“推送到问卷星”“创建在线问卷”“survey + interview guide”“convergent \u002F explanatory sequential \u002F exploratory sequential mixed methods instruments”。适用于社会科学、管理、教育、公共政策、医疗及其他经验研究；允许量化与质性部分使用相同或不同样本。产出研究设计判定、中央映射矩阵、问卷与变量字典、完整访谈三版、联合展示模板和预测试方案；默认用户成品仍是一份完整 Word。不要用于只设计单一访谈提纲（走 interview-guide-design）、只设计单一问卷、把现成工具做成网页、访谈转写、正式数据分析，或在没有研究问题的情况下机械拼接两套题目。","https:\u002F\u002Fgithub.com\u002Ffranklee16\u002Facademic-research-skills\u002Fblob\u002Fmaster\u002Frunning-surveys-experiments\u002Fmixed-methods-instrument-design-main\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Ffranklee16\u002Facademic-research-skills\u002Fblob\u002Fmaster\u002Frunning-surveys-experiments\u002Fmixed-methods-instrument-design-main\u002FSKILL.md",[214,215,216,217],{"categoryNo":131,"categoryName":132,"taxonomyType":22,"taxonomyTypeDesc":23},{"categoryNo":190,"categoryName":191,"taxonomyType":22,"taxonomyTypeDesc":23},{"categoryNo":161,"categoryName":162,"taxonomyType":22,"taxonomyTypeDesc":23},{"categoryNo":194,"categoryName":195,"taxonomyType":101,"taxonomyTypeDesc":102},[219,220,203,221,222,223,224,201,225,204,226,227,202],"methods","mixed methods instrument design","instrument","testing","数据","Word","survey","data analysis","工具",{"skillNo":229,"name":230,"nameZh":6,"description":231,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":232,"categories":233,"keywords":236,"publishedAt":45},"01M71FYX67NTZH9C05DSD325PD","scientific-critical-thinking","Evaluate scientific claims and evidence quality. Use for assessing experimental design validity, identifying biases and confounders, applying evidence grading frameworks (GRADE, Cochrane Risk of Bias), or teaching critical analysis. Best for understanding evidence quality, identifying flaws. For formal peer review writing use peer-review.","https:\u002F\u002Fgithub.com\u002Ffranklee16\u002Facademic-research-skills\u002Fblob\u002Fmaster\u002Freview-revise\u002Fscientific-critical-thinking\u002FSKILL.md",[234,235],{"categoryNo":190,"categoryName":191,"taxonomyType":22,"taxonomyTypeDesc":23},{"categoryNo":194,"categoryName":195,"taxonomyType":101,"taxonomyTypeDesc":102},[237,136,238,239,240,241,33,242,243,244,245,246,247,248,249,250,251,252,253,203],"critical","scientific critical thinking","thinking","write","frameworks","risk","evidence","grading","edit","read","peer review","claims","review","analysis","teaching","Bash","writing"]